An Architecture for Creating Collaborative Semantically Capable Scientific Data Sharing...

An Architecture for Creating Collaborative Semantically

Capable Scientific Data Sharing Infrastructures

Anuj R. Jaiswal, C. Lee Giles, Prasenjit Mitra, James Z. Wang

Presentation by Paulo Shakarian

Outline

• Problem• Overall Goal• Contributions• Metadata• Implementation• Future Work• Comparison to SIBDATA Concept

Problem

• Researchers often reference experimental results of their predecessors

• However, the raw data of experimental results is often not readily available.– Hence, results often cannot easily be re-used or

combined with other experiments

Problem (cont.)

• Large repositories (i.e. NASA, NOAA, etc.) do collect experimental data– Often conform to global schema (which may cause

some data to be lost)– Or stored as flat-files (requiring custom-built query

applications)• Also, data labels in experiments may differ (i.e.

Temp. vs. Temperature vs. Celsius)

Overall Goal

• Architecture for dissemination, sharing, querying, and searching of scientific data on the WWW

• Schema not known a-priori• Approach relies on sufficient meta-data of two

varieties:– Data about the experiment (conditions, source, when

uploaded, etc.)– Semantics for columns/rows in experimental results

(what they represent, what units, etc.)

Overall Goal (cont.)

• Two-part approach:– Annotation

application for semi-automatic creation of annotations

– Web-portal for searchable storage of annotated scientific data.

Contributions of the Paper

• Propose architecture for semantically capable collaborative infrastructure for data collection and sharing

• System that utilizes two-level metadata scheme for document description and dataset attributes

• Description of current implementation

Dataset Metadata

• Dublin Core (http://dublincore.org) is a set of 15 elements for minimal resource description to ensure minimal operability– OAI-PMH– IETF RFC 5013– ANSI/NISO Standard Z39.85-2007– ISO Standard 15836:2009

• Attributes listed on next 3 slides

Dataset Metadata

• Paper states “uses Dublin Core 15 elements” but actually uses the following 15:– Title– Creator– Subject– Description– Contributor– Publisher

– Date– Type– Format– Identifier– Source– Relation

– References– Is referenced by– Language– Rights– Coverage.

Attribute Metadata

• Challenges:– Same attribute, different row/column name– (i.e. Temp vs Temperature– Same row/column name, but different attribute (i.e. Temperature

(in deg C) vs Temperature (in deg K)– Row/column names may be ambiguous (i.e. Rate)

Attribute Metadata

• Metadata tags for attributes (right)

• Note they allow for dynamic generation of a dynamic collaboration ontology– Equivalent To– Different From– Superset Of– Subset Of– Type Of

Submitting a Dataset

• Uses a ``pull’’ technique– Author submits URL– System pulls annotated data

• Pull method allows the following– A moderator can check the URL from non-authorized

submitters– Automatic tagging of provenance information for

authorized users based on URL– Better protection from DOS attacks

• Banning of malicious users• Implement a round-robin policy for fetching

Implementation: Metadata• Used for chemical kinetics experiments• Experimental results in MS Excel• Metadata added through a MS Excel add-in

Implementation: Web Portal

• Three components– Web portal front-end– Data downloader and parser– Data analysis toolkit

Implementation: Web Portal

• Web Portal Front-End– Content management system– Dataset viewer– Data submission system

• Uses Mambo Server (open source, PHP-based) content-management system

• Data submission system deployed using JSP on ApacheTomcat 5

Implementation: Web Portal

• Data downloader and parser– Scheduler– Downloader– Parser

• Parser– Creates metadata as XML files– Data in Excel files imported into

MySQL database– Parser creates a dataset index,

linking dataset with dataset metadata and attribute metadata with data tables

Implementation: Data Analysis Tools

• In addition to supporting queries, plotting and regression tools included in web portal

Future Work

• Develop algorithms to derive dynamic collaboration ontology's

• Integrating query re-wrting and semantic searching using attribute-level semantics

• Automatic metadata generation using a user’s previous experiments

• Group, trust, privacy mechanisms

Comparison to SIBDATA Concept

• Relies on central repository (as opposed to multiple repositories for SIBDATA)

• Only useful for Excel-formatted experimental results

• Annotations may be an interesting feature to include in a SIBDATA or CDATA.

Questions